Certifiable Solver for Real-Time N-View Triangulation
Mercedes Garcia-Salguero, Javier González Jiménez
Abstract
Cutting-edge field robotic systems, such as UAV or autonomous cars, demand fast and optimal solutions for any component at the core of their critical navigational tasks. Among them, we focus on the triangulation of image points from multiple views, which is a cornerstone for more complex tasks such as visual localization and SLAM. In this letter we present a fast and certifiable solver for the N-view triangulation problem that doesn't require any specific optimization software package and can be implemented with any linear algebra library. The proposal relies on a series of linear convexifications which, in the limit, recovers the original problem, allowing us to solve problem instances with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$N=10$</tex-math></inline-formula> views in 150 microseconds on a standard desktop computer. On real data our solver obtains and certifies the optimal solution in more than 99% of the problem instances.
BibTeX
@inproceedings{ral2023_certifiablesolve,
title = {Certifiable Solver for Real-Time N-View Triangulation},
author = {Mercedes Garcia-Salguero and Javier González Jiménez},
booktitle = {RA-L 2023},
year = {2023}
}